System and method for inverse halftoning

1. A method for inverse halftoning, comprising the steps of: a) building a template training set including at least one continuous-tone image and a corresponding halftone image; b) retrieving a full set of templates and a set of corresponding output values from the template training set; c) representing each of the retrieved templates as t and applying a partitioning function to each of the templates t, the partitioning function dividing an initial look-up table into N smaller look-up tables, wherein N is an integer; d) storing each of the templates t and the corresponding output values for the templates in a corresponding one of the smaller look-up tables, the corresponding one of the smaller look-up tables having a numerical value associated therewith equal to a result of the partitioning function; e) setting the output value stored in the corresponding one of the smaller look-up tables equal to an averaged output value; and f) defining an integer r as the number of the retrieved templates and performing an inverse halftoning operation having the steps of: retrieving the r retrieved templates simultaneously; and inverse halftoning the r retrieved templates, the inverse halftoning using the N smaller look-up tables, the input to the inverse halftoning being a halftone image and the output of the inverse halftoning being a continuous-tone image.

2. The method for inverse halftoning as recited in claim 1 , wherein the step of applying the partitioning function comprises the steps of: a) establishing a variable p such that template t is a p-bit vector; b) dividing template t into log 2 N blocks, such that each said block except the last of said blocks has a width represented by size_of_blocks_except 13 last_block; c) establishing a function floor which rounds each said width to an integer that is less than or equal to [ p log 2 ⁢ N ] ; d) calculating each said width size_of 13 blocks 13 except 13 last_block as size_of ⁢ _blocks ⁢ _except ⁢ _last ⁢ _block = floor ⁡ [ p log 2 ⁢ N ] ⁢ bits ; e) calculating a width of the last of said blocks as size_of ⁢ _the ⁢ _last ⁢ _block = p - floor ⁡ [ p log 2 ⁢ N ] × ( log 2 ⁢ N - 1 ) ⁢ bits ; and f) establishing an intermediate set of variables a, b, i and j, and applying a bit-serial XOR operation to each said block independently as a(i)=t(j) t(j+1) . . . t(j+b), wherein i ranges between 0 and log 2 (N)−1, b =size_of_the_block_except_last_block when i <log 2 (N−1) and b=size_of _the_last _block when i=log 2 (N−1), and j = floor ⁡ [ p log 2 ⁢ N ] × i .

3. The method for inverse halftoning as recited in claim 2 , wherein N is selected as an integral power of 2, and the result of the partitioning function is given by result=a(0) ∥ a(1) ∥ . . . ∥ a(log 2 N−1).

4. The method for inverse halftoning as recited in claim 3 , wherein the partitioning function is applied to each of the r templates simultaneously, and if more than one of said r templates return the same result from the partitioning function, then a selected one of said templates is stored in the corresponding smaller look-up table and the remaining templates having the same result are discarded.

5. The method of inverse halftoning as recited in claim 4 , further comprising the steps of: a) appending the numbers 1 to r to the r retrieved templates; b) demultiplexing each template value to produce a demultiplexed value; c) decoding each demultiplexed value to produce a decoded value; d) multiplexing each demultiplexed value and the corresponding decoded value; e) ordering the respective output values of the templates with respect to the number assigned to each output value corresponding to the result of the partitioning function; and f) assigning output values to each discarded template equal to the output value of the left-most neighbor of each discarded template.